Discussion Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

[Vattila]

Except for the details about the improvements in the microarchitecture, we now know pretty well what to expect with Zen 3.

The leaked presentation by AMD Senior Manager Martin Hilgeman shows that EPYC 3 "Milan" will, as promised and expected, reuse the current platform (SP3), and the system architecture and packaging looks to be the same, with the same 9-die chiplet design and the same maximum core and thread-count (no SMT-4, contrary to rumour). The biggest change revealed so far is the enlargement of the compute complex from 4 cores to 8 cores, all sharing a larger L3 cache ("32+ MB", likely to double to 64 MB, I think).

Hilgeman's slides did also show that EPYC 4 "Genoa" is in the definition phase (or was at the time of the presentation in September, at least), and will come with a new platform (SP5), with new memory support (likely DDR5).

What else do you think we will see with Zen 4? PCI-Express 5 support? Increased core-count? 4-way SMT? New packaging (interposer, 2.5D, 3D)? Integrated memory on package (HBM)?

Vote in the poll and share your thoughts!

 

[jpiniero]

Greymon55 again :

https://twitter.com/x/status/1562751802059399168

76 23 1 19xx
77 23 1 20xx

R23 CB, 1T, 7600X ~1950 pts?
R23 CB, 1T, 7700X ~2050 pts?

If 7950X hits 5.8Ghz it should score ~2150 pts in R23 1T

So that's about what Alder Lake does. Wouldn't that be a little underwhelming, esp at similar clock speeds?

 

[Vope45]

So that's about what Alder Lake does. Wouldn't that be a little underwhelming, esp at similar clock speeds?

It's actually the reverse. CB favors intel and the lowest bin zen 4 core is already => highest bin gc core.

 

[inf64]

So that's about what Alder Lake does. Wouldn't that be a little underwhelming, esp at similar clock speeds?

Well we know that Golden Cove has around 10--11% higher IPC than Zen 3, and given what AMD supplied (+8-10% figure for Zen 4), there is no surprise. The IPC difference will be very small and app dependent (compute that is not AVX512 will favor intel, integer will *likely* favor AMD). It all comes down to ST and all core boost clocks. Oh, and yes, AMD will have Vcache models that help a lot with specific workloads.

 

[PJVol]

nvm

 

[Timorous]

So that's about what Alder Lake does. Wouldn't that be a little underwhelming, esp at similar clock speeds?

I expect CB to be on the low end of IPC improvements given that the L2 and DDR5 changes won't impact the score by much where as games will probably see a bigger boost due to those changes actually having an impact.

 

[PJVol]

Were there any leaks for 7600X CB MT score?

 

[Vope45]

I expect CB to be on the low end of IPC improvements given that the L2 and DDR5 changes won't impact the score by much where as games will probably see a bigger boost due to those changes actually having an impact.

This is correct. Zen 3 is only 11% faster than Zen 2 in cb whereas its actual uplift lies elsewhere. As I recall during zen 1 launch window, amd actually "convinced" maxon to update its suite. Can anyone confirm this ?

 

[Vope45]

Another one:

https://twitter.com/x/status/1562751758673137664

Stepping B1. Seems like a buggy launch to me.

 

[exquisitechar]

So that's about what Alder Lake does. Wouldn't that be a little underwhelming, esp at similar clock speeds?

It’s as expected based on everything we know so far. AMD has a slight PPC deficit compared to ADL and matches it/edges it out due to higher clocks. It’s only Cinebench, though. I suspect the situation may be more favorable for Zen 4 elsewhere.

I do find Raphael pretty underwhelming. Zen 4 will be hugely successful in servers and laptops, especially due to Intel’s terrible execution concerning the former, but it’s not very exciting in desktops and as a core. It seems that AMD is somewhat coasting by due to their node advantage this time. Still, I’m looking forward to seeing exactly how efficient it is.

 

[Asterox]

I expect CB to be on the low end of IPC improvements given that the L2 and DDR5 changes won't impact the score by much where as games will probably see a bigger boost due to those changes actually having an impact.

Cinebench is CPU rendering benchmark, and he dont care about CPU Cache size or sistem memory speed.

Same example, how much is Alder Lake IPC higher vs Zen 2, hm 25-30% higher IPC.

Page 33 - Zen 2 APUs/"Renoir" discussion thread

Page 33 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[deasd]

Greymon is hinting at 5.6Ghz all-core Turbo for 7600X?

https://twitter.com/x/status/1562707438885048320

IF this is a manual OC result, I would bet (but unconfident) that 7600X is still a TDP 65w SKU,

I thought the high TDP (105w) might eat up some OC headroom, but it doesn't looks like to be that case, instead it still has big headroom and could be insanely OC .

another situation is, 7600X would be still labeled TDP 105w but the PPT power would way less than 140watts (TDP*1.35, IIRC)



edit: it still has serious BIOS problem

https://twitter.com/x/status/1562772585003511808

https://twitter.com/greymon55/status/1562778222160330752

 

[Vattila]

I love how some people here can make seemingly ironclad conclusions about the perf/w of the 7950x

I think you mean calculations, if you refer to AMD's slide and the calculation done by @Abwx based on it. And you should love it, indeed. The slide clearly states >35% better performance and >25% better performance/Watt in Cinebench multithread, with further details given in endnotes. Calculating 7950X's score and power consumption in Watt is a simple exercise if you have the corresponding numbers for 5950X. Here are the slide and endnotes for your reference:

AMD CTO Mark Papermaster's PowerPoint Presentation at Financial Analyst Day 2022

 

[PJVol]

Something doesn't add up. 7600X is allowed to draw 50+% more of 5600X power, and is still gimped by 5.3Ghz ceiling?

 

[inf64]

Something doesn't add up. 7600X is allowed to draw 50+% more of 5600X power, and is still gimped by 5.3Ghz ceiling?

Maybe 7600X has dice that are lower quality and not able to reach higher all-core clocks @ stock without excessive Vcore?

 

[leoneazzurro]

Another one:

https://twitter.com/x/status/1562751758673137664

Stepping B1. Seems like a buggy launch to me.

There were several rumors giving the BIOS being tuned for compatibility and performance and thus the actual availability being postponed to the end of September. It's not unlikely that many ES out there are used with not final firmware. That said, there are so many variables in a result today, not forgetting proper cooling, RAM and power supply.

It’s as expected based on everything we know so far. AMD has a slight PPC deficit compared to ADL and matches it/edges it out due to higher clocks. It’s only Cinebench, though. I suspect the situation may be more favorable for Zen 4 elsewhere.

I do find Raphael pretty underwhelming. Zen 4 will be hugely successful in servers and laptops, especially due to Intel’s terrible execution concerning the former, but it’s not very exciting in desktops and as a core. It seems that AMD is somewhat coasting by due to their node advantage this time. Still, I’m looking forward to seeing exactly how efficient it is.

Is it really underwhelming? Probably some SKU are, if AMD wants to market them as direct competitors of RL, i.e. 7600Xvs13600, 7700X vs 13700. 7900X and 7950X already will be probably a different matter, however the card AMD will probably play is the efficiency one. For non-parallel tasks a 7600X and 7700X will be probably indistinguishable from a 13600/13700, while the power consumption is likely to be quite low (I don't see how people can think 7600X is not a 65W part). At that point they can attract some consumers anyway, as not everyone runs rendering works and in others application field, namely gaming, Zen4 and RL are likely to be very close.

 

[Zucker2k]

I think you mean calculations, if you refer to AMD's slide and the calculation done by @Abwx based on it. And you should love it, indeed. The slide clearly states >35% better performance and >25% better performance/Watt in Cinebench multithread, with further details given in endnotes. Here are the slide and endnotes for your reference:

View attachment 66582

AMD CTO Mark Papermaster's PowerPoint Presentation at Financial Analyst Day 2022

It says: overall performance improvement. So that should already include the power efficiency delta, which I suspect some of it is lost due to the high clocks Zen 4 will be running at. After all, the 25% figure was calculated at iso power. Moreover, I don't think AMD will be raising ppt to 230 watts, if it's getting 35% performance uplift at 25% less power. Something doesn't compute.

 

[uzzi38]

Another one:

https://twitter.com/x/status/1562751758673137664

Stepping B1. Seems like a buggy launch to me.

Might be software. I'm not aware of any bugs in a different software that utilises AVX-512

 

[inf64]

There were several rumors giving the BIOS being tuned for compatibility and performance and thus the actual availability being postponed to the end of September. It's not unlikely that many ES out there are used with not final firmware. That said, there are so many variables in a result today, not forgetting proper cooling, RAM and power supply.



Is it really underwhelming? Probably some SKU are, if AMD wants to market them as direct competitors of RL, i.e. 7600Xvs13600, 7700X vs 13700. 7900X and 7950X already will be probably a different matter, however the card AMD will probably play is the efficiency one. For non-parallel tasks a 7600X and 7700X will be probably indistinguishable from a 13600/13700, while the power consumption is likely to be quite low (I don't see how people can think 7600X is not a 65W part). At that point they can attract some consumers anyway, as not everyone runs rendering works and in others application field, namely gaming, Zen4 and RL are likely to be very close.

7700X (8C/16T) cannot be a competitor of 13700K (24T) part, it makes no sense. 7900X will compete with 13700K, 7700X will compete with 13600K and 7950X will compete with 13900K

 

[leoneazzurro]

7700X (8C/16T) cannot be a competitor of 13700K (24T) part, it makes no sense. 7900X will compete with 13700K, 7700X will compete with 13600K and 7950X will compete with 13900K

That is what I am saying, except some corner cases it will all depend on the market positioning of these SKUs. If AMD wants to sell the 7700K at the same price of the 13700K, then they will have a problem. If they will price them reasonably, then we will see a different situation. My point was exactly that in Zen4 case, there are no bad SKUs, there can be only bad pricing. The core seems to be definitely good in perf and perf/W figures. It would be interesting if we could see this generation 10-cores or 14- cores Zen4. 5900E is the proof that was definitely possible.

 

[Vattila]

It says: overall performance improvement. So that should already include the power efficiency delta

Yes, I didn't mean to imply that it did not, and I don't think @Abwx counted it twice either in his calculation. However, if you dispute his calculation, point out where he went wrong, and provide the correct one. If you have the score and wattage numbers for 5950X in the Cinebench multithread test, it should be a simple calculation to arrive at score and wattage for 7950X.

 

[inf64]

That is what I am saying, except some corner cases it will all depend on the market positioning of these SKUs. If AMD wants to sell the 7700K at the same price of the 13700K, then they will have a problem. If they will price them reasonably, then we will see a different situation. My point was exactly that in Zen4 case, there are no bad SKUs, there can be only bad pricing. The core seems to be definitely good in perf and perf/W figures. It would be interesting if we could see this generation 10-cores or 14- cores Zen4. 5900E is the proof that was definitely possible.

I see no way that AMD can ask the same price for 7700X (16T) as intel is asking for 13700K, the performance will not be there.

 

[nicalandia]

You haven't been keeping up with Raptor Lake leaks. Raptor Lake has a significantly improved memory subsystem and memory controller. You guys honestly think Intel was not going to improve the weakest part of Alder Lake? LOL!

I think you guys are in for a rude awakening.

Lets clear a few things here. CBR23 is not memory sensitive(as in cache subsystems) so any IPC gains there are purely raw performance gains.

As far as gaming and memory sensitive apps:

Zen3(with current L1D, L2 size) has better Cache Subsystems than Alder Lake and at ISO speed and ISO Memory they have the same gaming performance(See 12400F vs 5600X).

Yes Raptor Lake has an improved memory subsystem when compared to Alder Lake... But AMD also improved the Memory Subsystem on Zen4(50% Increase on L1/L2 DTLB entry and 100% increase on L2 size) and on top of that they have reached CPU Speed Parity and DDR5 Memory parity. And that will translate to higher gaming performance.

 

[leoneazzurro]

I see no way that AMD can ask the same price for 7700X (16T) as intel is asking for 13700K, the performance will not be there.

It is not a fact nor I believe that it will be a fact, I am responding to people who probably are suggesting that.
I was only suggesting that a part of the price premium some Intel SKU could keep over some other AMD Zen4 SKU may be offset by the lower power consumption.
Small example:
Hypotetical 13600K has a price of 300$, if linearly based on performance 7600X should be 200$, AMD may sell it at 230$ because of lower power consumption. But this is merely a hypothesis

 

[Abwx]

After all, the 25% figure was calculated at iso power. Moreover, I don't think AMD will be raising ppt to 230 watts, if it's getting 35% performance uplift at 25% less power. Something doesn't compute.

That s not at isopower, the CPU provide >35% better perf at a given power such that it has >25% better perf/watt than a 5950X.

At equal perf/watt it would consume 1.35 x 142 = 191.7W

But since it has 25% better perf/watt it consume 191.7/1.25 = 153.36W.

 

[PJVol]

Maybe 7600X has dice that are lower quality and not able to reach higher all-core clocks @ stock without excessive Vcore?

Yep, kinda like in a recent 5600, 5700x.
My office 5700g has similar cores.